0121 448 4776

4.7

★★★★★

Clutch

I built my product with AI tools. Now what?

You got further in a weekend than you thought possible. Now it’s stuck at 80%, every prompt breaks something else, and you’re not sure whether to keep going, fix it or start again.

The short version

  • AI is brilliant at the obvious, and a prototype is the obvious bit.
  • The problems aren’t on the screen. They’re underneath it, where nobody looked.
  • More of the same usually makes it worse.
  • The honest answer depends on what’s underneath, not on how it looks.
  • Decisions first, then code that someone can live with.
  • Find out what you’ve got before you decide what to do with it.

If you’ve got an app built with AI tools like Lovable, Bolt or v0, you’ve probably followed the same path as everyone else. The first few days felt like magic. You described what you wanted and there it was, on screen, working.

Then you tried to add logins, or payments, or connect it to the system your business actually runs on. That’s where the magic stopped.

This isn’t a criticism of you or the tools. It’s what they’re built to do.

Why it worked at first

AI is brilliant at the obvious, and a prototype is the obvious bit.

A login screen, a dashboard, a form, a list of things. These have been built millions of times, so the tools have seen millions of examples. Ask for one and you get a good, average version.

The parts that make your product yours are different. How your pricing works, who’s allowed to see what, what happens when a customer does something odd. There’s no average answer to copy, so the tool guesses. Guesses look fine on screen. They show up later.

That’s the 80% wall. The tool didn’t get worse. You reached the part that needs decisions, not code.

What usually breaks in an app built with AI tools

The problems aren’t on the screen. They’re underneath it, where nobody looked.

Most of what goes wrong only shows up when real people, real data or real money arrive.

  • Security is a guess. The logins work, but nobody has checked who can see whose data. Often the answer is anyone who knows where to look.
  • The data has no shape. Information is stored however the tool decided on the day, so reports, exports and the next feature all have to fight it.
  • Payments and integrations half work. The straightforward case is fine. A failed card, a timeout or another system that doesn’t answer, and it falls over, often without telling anyone.
  • Nobody can change it safely. There’s no structure a developer recognises, so every fix risks breaking something two screens away.
  • Every prompt costs credits and confidence. You fix one thing, two others break, and you’re going round in circles paying for the privilege.
  • It looks like everyone else’s. Same layout, same buttons, same colours. Fine for a test. Not for something you’re asking people to trust, or pay for.

What doesn’t fix it

More of the same usually makes it worse.

“One more prompt will sort it.” If the last twenty didn’t, the twenty-first probably won’t. The tool can’t see the problem, because it made it.

“A different AI tool will do better.” It’ll make different guesses. The decisions still haven’t been made.

“Let’s throw it away and start again.” Sometimes that’s right. Often it’s wasteful, because plenty of AI-built products have parts worth keeping.

“We just need a cheap developer to patch it.” Patching without looking at the whole thing is how you end up with a product nobody can change.

Fix, extend or rebuild

The honest answer depends on what’s underneath, not on how it looks.

There are three routes, and most products fit one of them clearly once someone has looked.

  • Fix it if the core works, the data makes sense and the problems are in specific places: a security gap, a payment flow, screens that break on mobile.
  • Extend it if the foundations are sound and you need what the tool couldn’t do: proper logins and permissions, connections to your CRM or finance system, an admin side your team can actually use.
  • Rebuild it if the tool built something different from what you need, or the code is so tangled that every change is a gamble. Keep the learning, the screens that tested well and the idea. Let the code go.

Most people can’t tell which one they’re looking at, because they can’t see underneath. That’s not a skills gap. It’s what you’d expect when the tool wrote the code and you wrote the prompts.

What a proper rebuild involves

Decisions first, then code that someone can live with.

Whether it’s a fix, an extension or a full rebuild, the work has the same shape. Decide what the product has to do and who it’s for. Then build it so it holds up when real people use it. That’s what our web applications and integrations service does, quoted per project.

We use AI tools in our own builds. They let us build better things than we could before. What we bring is the part they can’t: knowing what’s worth building, and what will break.

  • Decisions written down. Who it’s for, what it must do and what it mustn’t. The part a prompt never captured.
  • Security and data done properly. Who can see what is designed, not guessed, and the data has a structure the next feature can use.
  • Built to fail safely. Every connection has a plan for the day the other end stops answering.
  • Designed around the people using it. It’s why the schools platform we built in 2017 is still used every day by more than 5,000 schools.
  • You own it. Documented, handed over, no lock-in, and no dependency on us to keep it running.

Where to start

Find out what you’ve got before you decide what to do with it.

The UX audit Clarity Review is built for early-stage products and AI-built tools. It’s £550 + VAT, delivered in three working days. We look at what the product is, how people move through it and where they get stuck, and you get a clear report of what to fix first.

The report is yours. Give it to your own developer if you like. If you want us to do the work, the next step is a conversation about fixing, extending or rebuilding, quoted before anyone writes code.

And if nobody has agreed what the product should do yet, settle that first. A one-day strategy workshop, from £2,800 + VAT, gets the people who matter to decide.

Questions about apps built with AI tools

Stop prompting and find out what’s underneath. Then decide whether it needs fixing, extending or rebuilding. A Clarity Review, £550 + VAT, tells you which in three working days.

Often it can be fixed or extended. We only suggest a rebuild when the foundations won’t hold what you need, and we’ll tell you early.

When the tool built something different from what you need, or when every change breaks something else. Keep what you learned and the screens that worked, and rebuild the rest properly.

It’s quoted per project, because fixing a payment flow and rebuilding with logins and integrations are very different jobs. Start with the audit and you’ll know which end of the range you’re at.

Free resource

3 reasons your website isn’t converting. The three things that usually stop people getting in touch, and how to spot them in your own product.

The one thing to do next

Web applications and integrations

Stuck at 80%? We’ll tell you whether to fix, extend or rebuild, then build it so it holds up. Start with a Clarity Review, £550 + VAT.

The one thing to do next

Settle the story in a day.

Your team in a room, your audiences and value pillars on the wall, and the words everyone uses from then on. From £2,800.